@aliasmode/anyfetch 0.8.0 → 0.8.2
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +17 -1
- package/package.json +6 -6
package/README.md
CHANGED
|
@@ -179,7 +179,7 @@ Anyfetch is in the official [MCP Registry](https://registry.modelcontextprotocol
|
|
|
179
179
|
- Read pages: `anyfetch fetch <url> [<url>...] --format markdown`
|
|
180
180
|
```
|
|
181
181
|
|
|
182
|
-
**Make it the default.**
|
|
182
|
+
**Make it the default.** Anyfetch tells the agent to use its tools in place of the built-in web tools. In our test with Claude Code, the agent then chose Anyfetch for 8 of 8 web tasks. Before, it chose Anyfetch for 2 of 8. To take the choice away, turn the built-ins off:
|
|
183
183
|
|
|
184
184
|
- Claude Code, in `~/.claude/settings.json`: `"permissions": { "deny": ["WebFetch", "WebSearch"] }`
|
|
185
185
|
- Codex, in `~/.codex/config.toml`: `web_search = "disabled"`
|
|
@@ -353,6 +353,22 @@ In single runs, Anyfetch answered 32 and 33 questions, and the built-in tools 28
|
|
|
353
353
|
|
|
354
354
|
Three things to know about this test. The question writer used Anyfetch to check each answer, so the first group has only pages that Anyfetch can read. The Financial Times was on the writer's list and has no question: its articles need a subscription. Thirty of the questions are one day older than the test; 28 of their answers were checked on the source page again that day. Every answer is in [`bench/results/tasks.tsv`](bench/results/tasks.tsv).
|
|
355
355
|
|
|
356
|
+
We also read the same 97 sites with eight other tools, each with its default settings and no API key:
|
|
357
|
+
|
|
358
|
+
| Tool | Pages read |
|
|
359
|
+
| --- | --- |
|
|
360
|
+
| **Anyfetch** | **82** |
|
|
361
|
+
| Exa MCP (hosted, answers from stored copies) | 78 |
|
|
362
|
+
| Scrapling, stealth browser | 71 |
|
|
363
|
+
| Scrapling, plain HTTP | 69 |
|
|
364
|
+
| Hermes Agent `web_extract` (hosted readers) | 68 |
|
|
365
|
+
| Crawl4AI | 66 |
|
|
366
|
+
| Jina Reader | 61 |
|
|
367
|
+
| Playwright MCP | 52 |
|
|
368
|
+
| DuckDuckGo MCP server | 49 |
|
|
369
|
+
|
|
370
|
+
How each one ran, and what to know before you compare, is in [`bench/other-tools/NOTES.md`](bench/other-tools/NOTES.md). The full ranking with every site is on [aliasmode.com](https://aliasmode.com/anyfetch/best-web-fetch-tools-for-ai-agents/).
|
|
371
|
+
|
|
356
372
|
The per-site results are in [`bench/results/`](bench/results). Sites change their rules often, and results differ between networks, so run it yourself:
|
|
357
373
|
|
|
358
374
|
```sh
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@aliasmode/anyfetch",
|
|
3
|
-
"version": "0.8.
|
|
3
|
+
"version": "0.8.2",
|
|
4
4
|
"description": "Web search and web fetch for AI agents, with no API keys. Reads the sites that refuse AI agents.",
|
|
5
5
|
"keywords": ["mcp", "mcp-server", "web-search", "web-fetch", "ai-agents", "claude-code", "codex", "cursor"],
|
|
6
6
|
"license": "Apache-2.0",
|
|
@@ -11,10 +11,10 @@
|
|
|
11
11
|
"files": ["run.js"],
|
|
12
12
|
"engines": { "node": ">=16" },
|
|
13
13
|
"optionalDependencies": {
|
|
14
|
-
"@aliasmode/anyfetch-darwin-arm64": "0.8.
|
|
15
|
-
"@aliasmode/anyfetch-darwin-x64": "0.8.
|
|
16
|
-
"@aliasmode/anyfetch-linux-arm64": "0.8.
|
|
17
|
-
"@aliasmode/anyfetch-linux-x64": "0.8.
|
|
18
|
-
"@aliasmode/anyfetch-win32-x64": "0.8.
|
|
14
|
+
"@aliasmode/anyfetch-darwin-arm64": "0.8.2",
|
|
15
|
+
"@aliasmode/anyfetch-darwin-x64": "0.8.2",
|
|
16
|
+
"@aliasmode/anyfetch-linux-arm64": "0.8.2",
|
|
17
|
+
"@aliasmode/anyfetch-linux-x64": "0.8.2",
|
|
18
|
+
"@aliasmode/anyfetch-win32-x64": "0.8.2"
|
|
19
19
|
}
|
|
20
20
|
}
|